> [...] there is no practical need for floating point / double or quad
> precision. [...]
This is not really correct IMHO. There *are* parts in FreeType that
need high-precision computation internally (SDF is one example IIRC,
another one is font variation with higher-order interpolation); using
floating point arithmetic would certainly be helpful and simplify the
code – and probably even be faster.
On the other hand, FreeType is an old library. In earlier times,
implementations of IEEE floating point arithmetic were buggy for a
surprisingly high number of platforms and/or compilers; this justified
the use of fixed-point arithmetic then. As a beneficial side effect,
the current code compiles almost everywhere, for almost every
processor. I don't see a compelling reason to change that.
If somebody is going to develop FreeType 3, with a shiny new API,
supporting full thread-safety, and probably written in another
programming language (you name the new kid on the block :-), then
floating-point arithmetic is the way to go, I think.
Werner
